[+]
Getting Started
[+]
Recognition
[-]
Detection
  
 [-]
Detection Documentation
   
   
   
   
   
[+]
Segmentation
[+]
Resources
Updated on 9/11/2019
Brain Builder Knowledge Base
Exporting Detection Data
Direct link to topic in this publication:
  • Detection
  • »
  • Detection Documentation
  • »
  • Exporting Detection Data
In This Section:

Exporting Data

After you have annotated data in Brain Builder, you may want to export it so that you or someone else can use it to train a DNN. Brain Builder provides a number of Data Export options so that you can export exactly the data you want in the format you need.

  1. Start the export process from one of the following locations:

    • Dashboard — From the Dashboard, click the Menu button on the project you wish to export and select Export Project.

    • Project page — From the Project page, click the Export Project button next to the name of the project.

    • Dataset page — From the Dataset page, click the Export Dataset button next to the name of the dataset.

    • Exports page — Navigate to the Exports page from the User menu in the top-right corner, and click the + Export button at the top-right of the page.

  2. Select a framework.

    Neurala can export data in one of three formats:

    If you are unsure of which export format you need, consult the person or team who will be working with the data.

    • TensorFlow—The data is structured to work directly in the TensorFlow AI framework.

    • Caffe—The data is structured to work directly in the Caffe AI framework.

    • JSON—The data is exported in structured text format for easy adaptation to any purpose.

  3. Select an Export Type (for TensorFlow or Caffe exports).

    • Image Segmentation exports pixel-level masks showing where the objects have been tagged.

    • Object Detection exports rectangular bounding boxes.

  4. Select which project(s) you want to include in the export. Only images and classes included in these projects will be exported.

  5. Select which class(es) in the project you want to include in the export.

    Selecting the Customize my Palette box gives you the opportunity to further customize the way classes are exported. It allows you to drag the classes to change their order in the export, including the indexed PNG masks, and to use the Alias field to combine multiple classes under the same new class label.

  6. Select the Export Options (for TensorFlow or Caffe exports). If you are unsure about which of these options to select for your project, consult with the person who will be using the data.

  • Normalize Bounding Box Coordinates — By default, bounding box coordinates are exported as absolute pixel values. When selecting this option, coordinates are exported in normalized fractions from 0 to 1.
  • Resize Images — Set a standard height and width for all images. Images and masks will be resized to these dimensions, which may result in a skewed appearance.
  • Include Images — Choose whether or not to export the images in this export.
  • Dataset Allocation — Define the percentage of images you want separated into the Training, Validation, and Test subsets. The default is 80/10/10, which is common among AI practices.

Then, click Review and confirm that the Export Settings meet your expectations.

Leave the checkbox selected if you want to receive an email when the export is ready to be downloaded.

So What Exactly Gets Exported?

For every frame, Brain Builder exports the relevant polygons that are shown on the Workspace.

For example, in the screenshot below, you can see a track on the timeline for every time you have run the video annotator. The tracks are displayed from most recent (top) to least recent (bottom). Brain Builder displays and exports the most recent (top) track on your timeline for any given frame. However, if ground truth is ever manually entered—i.e. a polygon is created manually by a human—then that polygon is what gets exported regardless of what tracks exist.

Next: Workspace Hot Keys >>